Bringing People Along: Up-skilling, Skeptics, and the Fear Nobody Names

Start with the numbers, because they reframe the skeptic from an attitude problem into a majority position, especially outside the tech hubs you think of when you envision the “AI-native designer.”

Pew’s survey of 5,273 U.S. workers found 52% worried about the future impact of AI at work, against 36% hopeful, with a third feeling overwhelmed outright and 32% expecting AI to mean fewer job opportunities for them, personally, long-term. And the fear doesn’t just sit there; it drives behavior you can’t see. Microsoft and LinkedIn’s Work Trend Index found 52% of AI users reluctant to admit using it on their most important work.

Ethan Mollick’s secret cyborgs (the hidden adopters I leaned on in the opening piece of this series) are one face of that fear. The quiet skeptic on your team is the other face of the same fear. One hides their usage; one hides their dread. Both are managing risk you haven’t made it safe to surface.

Which is why the foundational text for AI adoption isn’t an AI text at all. It’s Amy Edmondson’s research on psychological safety:

If you want to see what happens when an org introduces a capability leap and ignores the human learning system around it, Matt Beane’s study of robotic surgery captures the up-skilling gap: 18 top U.S. teaching hospitals, watching what the da Vinci robot did to how residents learn.

In open surgery, residents learned by doing: hands in the field, graduated responsibility, the senior surgeon physically unable to do it all alone. The robot broke that contract. One surgeon could now do the whole procedure; the resident stood at a console or watched a screen, occasionally seized-back controls, learning almost nothing. As Beane put it:

The standard up-skilling failure is treating adoption as a single population with a single lever. But the stance people take toward this technology isn’t one variable, and the data above says the stances aren’t even about the technology.

Here’s what I suggest: a tiered plan that meets each stance with the move it actually needs. Map your roster to it privately, with your leads. These are diagnostic states people move through, not castes you create.

Two notes on the table.

• The tiers are not a maturity ladder. A Guardian is not a Pragmatist who needs more convincing. “Skepticism about quality” and “fear about identity” are different feelings needing opposite responses, and next up in this series is a whole piece on telling them apart.
• Notice that not one tier’s “move” is a training course. Courses are fine. But they’re just never the binding constraint.

Handling the up-skilling performance conversation with your strong, traditionally skilled designer starts with empathy and ends with a plan.

1. Open with observation, not diagnosis. “I’ve noticed you’ve been quiet in the show-and-tells, and I don’t think it’s disinterest. I’m not here to pitch you on any tool — I want to know where you actually are with all this.”

   • You are signaling that the meeting is safe before asking them to take a risk in it.

2. Name the fear yourself, so they don’t have to go first. “Half of everyone out there is worried about what AI does to their job. If some version of that is in the room for you, it’s a rational thing to have in the room. Saying it here costs you nothing — I mean that literally: nothing you say in this conversation touches your review.”

   • You cannot ask for interpersonal risk-taking at a moment of maximum threat without underwriting it explicitly.

3. Separate their worth from their output. “You were never valuable to me because you produce screens quickly. You’re valuable because you know when a design is wrong, and why, before anyone else in the room does. That skill got more valuable this year, not less — production got cheap, which makes judgment the scarce thing we pay for.”

   • This is the judgment-column argument delivered to one human, and it only lands if your leveling rubric actually backs it up. If it doesn’t yet, say that honestly too.

4. Be honest about what is changing. No false promises; they will smell one instantly, and a broken promise here is unrecoverable. “I can’t tell you the job stays the same, because it won’t. The production parts are getting automated, and pretending otherwise would be me managing you badly. What I can tell you is what this org is doing about it, recalibrating what it rewards, and that I’d rather run that transition with you than around you.”

5. Make one small, self-chosen ask. “Pick one task you already do: Your choice, not mine, and run it once with the AI workflow. I want your read on where it fails, because you’ll see failures the enthusiasts won’t.”

   • Note what this does: it routes their first contact through their strength (critical judgment), not their deficit. For a threatened person, “help me find what’s wrong with it” is a fundamentally different invitation than “catch up.”

6. Commit to something yourself, in writing, before they leave. Credit for anything they surface. Protected practice time on the calendar. And the evaluation commitment: “This cycle, you will not be scored on adoption. You’ll be scored on the same thing as always — critical thinking and judgment. Including judgment about these tools and processes.”

   • A commitment that costs you nothing is read, correctly, as worth nothing. This one costs you a real position in calibration, which is why it works.

What the conversation never does: promise safety it can’t deliver, argue about the technology’s merits, or end with a deadline. The conversation’s job is to convert an unspeakable fear into a discussable career question.

1. Week one: map the team. Keep it casual and nonjudgmental. Leads will want a fifth column for “stubborn/hopeless.” Refuse it. The moment a tier becomes terminal, the mapping leaks and becomes a purge list, and per the Pew numbers, you might be purging half the team.
2. Run the amnesty if you haven’t. It’s step two of the ops piece — because the tiered plan runs on real data about who’s actually where, and you don’t have that while usage is hidden.
3. Give every Guardian a quality job within the fortnight. The eval and quality work your org now owns is the single best container for skepticism ever invented: it pays the skeptic to be right. Your loudest skeptic may be a worse first recruit than your quietest one. The loud one has a public position to defend; the quiet one just has standards.
4. Schedule the identity conversations, threatened tier first, and rehearse. Managers should run the script once with each other before running it live. The wrong way is a manager who reads a script off a card and then flinches when it works and a real fear lands on the table.
   • This could be slow. Six conversations is a manager’s whole week of 1:1 capacity. It is still faster than the alternative, which is eighteen more months of a stalled program plus attrition of exactly the people with the deepest craft.
5. Protect practice time where everyone can see it. Build open learning forums, and emphasize a learning mindset with your team. Beane’s residents lost their learning to efficiency pressure one reasonable-sounding decision at a time. Practice and learning blocks on the calendar have to be protected by leads, lest sprint pressure kills trust that this is a real thing.
6. Measure movement. Quarterly: how many people shifted tiers, how many library contributions came from former Guardians and threatened-tier folks, how many identity conversations have actually happened. Not measured: monthly AI-sentiment surveys (they teach people performative enthusiasm — the exact opposite of Amy Edmondson’s variable).

• U.S. Workers Are More Worried Than Hopeful About AI — Pew Research Center — the numbers that make fear a majority position. For your next leadership conversation about “resistors.”
• Psychological Safety and Learning Behavior in Work Teams — Amy Edmondson, ASQ 1999 and The Fearless Organization. The mechanism connecting safety to learning.
• Learning to Work with Intelligent Machines — Matt Beane, HBR (and the open-access UCSB summary of the ASQ shadow-learning study) — the robotic-surgery natural experiment. Hand it to whoever controls sprint capacity.
• Detecting the Secret Cyborgs — Ethan Mollick — hidden adoption as a fear behavior, and the incentive design that reverses it.
• Microsoft & LinkedIn 2024 Work Trend Index — the 52%-hide-their-usage stat. Shadow-IT behavior at population scale.
• Stop Making Your Team Figure Out AI on Their Own — Laura Klein, NN/g — the ops-owns-enablement argument. Pairs with the tier table’s “your move” column.

Back to the fear nobody says in training:
if the machine can do the part of my job I love, what am I?

That question gets answered one conversation at a time, by a manager willing to be honest:

The drawing rectangles, pixel-pushing part is changing. The “knowing what’s good” part is the job now, and we’re rebuilding the org’s rewards around that on purpose, with you — if you’ll come.

Fear that can’t be spoken becomes resistance.
Fear that can be spoken becomes a career conversation.

Have the conversation.